METHODOLOGY

How SIGNAL produces intelligence briefings

Sources, scoring methodology, AI transparency, and what we do not do.

SOURCE SELECTION

SIGNAL monitors dozens of credentialed sources: defense and foreign-policy think tanks, OSINT researchers, and national-security analysts. Every source has a documented, public track record. Sources include RAND, CSIS, the Carnegie Endowment, Brookings, the Atlantic Council, Bellingcat, Lawfare, War on the Rocks, and the Cipher Brief.

NATO/ADMIRALTY CODE

Every source is rated using the NATO Admiralty Code (AJP-2.1, STANAG 2511). Source reliability and information credibility are assessed independently.

RATING	RELIABILITY	EXAMPLE
A	Completely reliable	Primary government document
B	Usually reliable	Major think tank
C	Fairly reliable	Specialist blog with track record
D	Not usually reliable	Anonymous channel
E	Unreliable	Known inaccurate content
F	Cannot be judged	New source

EPISTEMIC STATUS TAGS

CONFIRMEDVerified by 3+ independent reliable sources

LIKELYSupported by 2 independent sources

ASSESSEDAnalytical judgment from patterns and context

REPORTEDSingle source, unverified — always labeled

SPECULATIVEInference with low evidence base

UNVERIFIEDCannot assess reliability

AI TRANSPARENCY

SIGNAL uses Claude (Anthropic) for summarization and briefing synthesis. Every briefing undergoes human editorial review before publication. The AI is instructed to only use information from source texts, cite every claim, use hedging language, and state “source does not specify” rather than inferring. All published briefings carry a visible “AI-curated analysis” label and machine-readable metadata per EU AI Act Article 50.

WHAT WE DO NOT DO

We never publish trust scores for named individuals. We never reproduce verbatim article text. We never monitor X/Twitter, Reddit, or YouTube transcripts. Raw text is deleted after 7 days. Only AI-generated summaries are retained.